Laser & Optoelectronics Progress, Volume. 61, Issue 4, 0437008(2024)

Underwater Image Enhancement Based on Multi-Scale Attention and Contrast Learning

Yue Wang1,2,3, Huijie Fan1,2、*, Shiben Liu1,2,3, and Yandong Tang1,2
Author Affiliations
  • 1State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, Liaoning, China
  • 2Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang 110016, Liaoning, China
  • 3University of Chinese Academy of Sciences, Beijing 100049, China
  • show less
    Figures & Tables(11)
    Overall network structure
    Multiscale channel pixel attention module
    Feature matching experimental results. (a) Original image; (b) Fusion-based; (c) Water-Net; (d) Ucolor; (e) ours
    Results of the standard color card recovery experiments. (a) Original image; (b) ULAP; (c) FUNIE-GAN; (d) Water-Net; (e) Shallow-UWnet; (f) Ucolor; (g) ours; (h) standard color card
    Enhancement results of the different methods on UIEBD. (a) Original image; (b) UDCP; (c) Fusion-based; (d) ULAP; (e) Water-Net; (f) UWCNN; (g) FUNIE-GAN; (h) Shallow-UWnet; (i) Ucolor; (j) ours; (k) GT
    Enhancement results of the different methods on EUVP1 and EUVP2. (a) Original image; (b) UDCP; (c) Fusion-based; (d) ULAP; (e)Water-Net; (f) UWCNN; (g) FUNIE-GAN; (h) Shallow-UWnet; (i) Ucolor; (j) ours; (k) GT
    Enhancement results of the different methods on RUIE and HRCID . (a) Original image; (b) Water-Net; (c) UWCNN; (d) FUNIE-GAN; (e) Shallow-UWnet; (f) Ucolor; (g) ours
    • Table 1. PSNR and SSIM values for the different methods on UIEBD, EUVP1 and EUVP2 datasets

      View table

      Table 1. PSNR and SSIM values for the different methods on UIEBD, EUVP1 and EUVP2 datasets

      MethodUIEBDEUVP1EUVP2
      PSNRSSIMPSNRSSIMPSNRSSIM
      UDCP511.130.481413.570.522212.310.3355
      Fusion-based420.660.816017.540.600319.530.5922
      ULAP716.630.604819.810.733417.140.5046
      Water-Net921.100.868918.700.724020.880.7168
      UWCNN2712.930.442516.420.590717.880.6588
      FUNIE-GAN1419.170.772824.380.870422.620.7250
      Shallow-UWnet1118.270.707223.860.837522.400.6892
      Ucolor1020.690.803123.960.860723.780.7699
      Ours23.830.886625.460.894624.390.7952
    • Table 2. Number of feature point matches for the different methods

      View table

      Table 2. Number of feature point matches for the different methods

      Image No.Original imageFusion-based4Water-Net9Ucolor10Ours
      Average17.2524.2529.5026.5041.75
      1916232139
      23229323546
      332120626
      42531434456
    • Table 3. UCIQE and entropy values for the different methods on RUIE and HRCID datasets

      View table

      Table 3. UCIQE and entropy values for the different methods on RUIE and HRCID datasets

      MethodRUIEHRCID
      UCIQEEntropyUCIQEEntropy
      Water-Net90.56807.50920.52077.4918
      UWCNN270.44817.24630.45546.9498
      FUNIE-GAN140.55357.49840.55797.4125
      Shallow-UWnet110.38356.65030.38726.3581
      Ucolor100.52587.36530.47667.2015
      Ours0.58027.66680.54877.6461
    • Table 4. Results of the ablation experiments

      View table

      Table 4. Results of the ablation experiments

      ModelPSNRSSIM
      Model 123.130.8711
      Model 223.160.8824
      Model 2(add Conv 7×7)23.690.8847
      Model 323.490.8835
      Model 3(add Conv 5×5)23.510.8845
      Model 423.390.8845
      Model 4(add Conv 3×3)23.560.8846
      Model 523.450.8796
      Model 623.300.8783
      Ours23.830.8866
    Tools

    Get Citation

    Copy Citation Text

    Yue Wang, Huijie Fan, Shiben Liu, Yandong Tang. Underwater Image Enhancement Based on Multi-Scale Attention and Contrast Learning[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0437008

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Digital Image Processing

    Received: Oct. 17, 2022

    Accepted: Dec. 7, 2022

    Published Online: Feb. 26, 2024

    The Author Email: Huijie Fan (fanhuijie@sia.cn)

    DOI:10.3788/LOP223047

    CSTR:32186.14.LOP223047

    Topics